Re: [PATCH 0/6] thread lazy_init_name_hash

From: Jeff Hostetler <git@jeffhostetler.com>
Date: Mar 22, 2017, 19:22 UTC
Message-ID: <90a7b4fa-67bf-6501-1da6-97883d7cde4a@jeffhostetler.com>
In-Reply-To: <CAGZ79kYh0eP-dzpX58SrSRJGEr9iyOb0Q3vizsvxE81kzddwxQ@mail.gmail.com>

On 3/22/2017 2:02 PM, Stefan Beller wrote:

Show 9 quoted lines

> On Wed, Mar 22, 2017 at 10:14 AM,  <git@jeffhostetler.com> wrote:
>>
>> During our testing on the Windows source tree (3.1M
>> files, 500K folders, 450MB index), this change reduced
>> the runtime of lazy_init_name_hash() from 1.4 to 0.27
>> seconds.
>
> This sounds promising. :)
> A fast skim over the code makes me like the code.

I forgot to mention that this was on an 8 (logical) core machine.

Show 6 quoted lines

>
>>  hashmap.c                           |  29 ++-
>>  hashmap.h                           |  25 ++
>
> Could you add some documentation to
> Documentation/technical/api-hashmap.txt ?

Will do. Thanks.

> (Bonus points for migrating the documentation inline,
> c.f. discussion surrounding [1])
>
> [1] https://public-inbox.org/git/20141212212800.GA27451@peff.net/

I'd like to save this for a separate effort.

Show 7 quoted lines

>
>>  name-hash.c                         | 490 +++++++++++++++++++++++++++++++++++-
>
> AFAICT the new threading is all implicit in name-hash and we do not expose
> its functionality or tuning knobs outside the testing helper, such that we do
> not need API documentation here, but only enough code comments to
> understand the code for maintainability?

Yes, my goal was to just make it transparent to all callers. Especially since the main lazy_init_name_hash() function is static and not visible anyway.

Jeff